Global sensitivity analysis for a numerical model of radionuclide migration from the RRC “Kurchatov Institute” radwaste disposal site

Abstract

Today, in different countries, there exist sites with contaminated groundwater formed as a result of inappropriate handling or disposal of hazardous materials or wastes. Numerical modeling of such sites is an important tool for a correct prediction of contamination plume spreading and an assessment of environmental risks associated with the site. Many uncertainties are associated with a part of the parameters and the initial conditions of such environmental numerical models. Statistical techniques are useful to deal with these uncertainties. This paper describes the methods of uncertainty propagation and global sensitivity analysis that are applied to a numerical model of radionuclide migration in a sandy aquifer in the area of the RRC ''Kurchatov Institute'' radwaste disposal site in Moscow, Russia. We consider 20 uncertain input parameters of the model and 20 output variables (contaminant concentration in the observation wells predicted by the model for the end of 2010). Monte Carlo simulations allow calculating uncertainty in the output values and analyzing the linearity and the monotony of the relations between input and output variables. For the non monotonic relations, sensitivity analyses are classically done with the Sobol sensitivity indices. The originality of this study is the use of modern surrogate models (called response surfaces), the boosting regression trees, constructed for each output variable, to calculate the Sobol indices by the Monte Carlo method. It is thus shown that the most influential parameters of the model are distribution coefficients and infiltration rate in the zone of strong pipe leaks on the site. Improvement of these parameters would considerably reduce the model prediction uncertainty.

An efficient methodology for modeling complex computer codes with Gaussian processes

“…Only the correlation coefficients vector θ has to be estimated. Then, sensitivity measures such as the Sobol indices (Saltelli et al (2000), Volkova et al (2008)) are computed and used to sort the inputs by influence order.…”

“…⊲ Boosting of regression trees: this model was used in the previous study of the data (Volkova et al (2008)). The boosting trees method is based on sequential construction of weak models (here regression trees with low interaction depth), that are then aggregated.…”

Global sensitivity analysis with 2d hydraulic codes: applied protocol and practical tool

“…The first two steps lead to define probability density function constructed to represent uncertainty of selected input parameters for the study. It has to be emphasized that this step of the GSA process is important and strongly subjective [Volkova et al, 2008]. Propagation of uncertainty is then required (step ii of the UA), all sources of uncertainties are varied simultaneously, which is classically done using Monte-Carlo techniques or more parsimonious Monte-Carlo like approach [Iooss, 2011].…”

How to effectively compute the reliability of a thermal–hydraulic nuclear passive system

“…In [Bucher and Most, 2008;Gavin and Yau, 2008;Liel et al, 2009], polynomial Response Surfaces (RSs) are employed to evaluate the failure probability of structural systems; in Fong et al, 2009;Mathews et al, 2009], linear and quadratic polynomial RSs are employed for performing the reliability analysis of T-H passive systems in advanced nuclear reactors; in [Deng, 2006;Hurtado, 2007;Cardoso et al, 2008;Cheng et al, 2008], learning statistical models such as Artificial Neural Networks (ANNs), Radial Basis Functions (RBFs) and Support Vector Machines (SVMs) are trained to provide local approximations of the failure domain in structural reliability problems; in [Volkova et al, 2008;Marrel et al, 2009], Gaussian meta-models are built to calculate global sensitivity indices for a complex hydrogeological model simulating radionuclide transport in groundwater.…”